The thread-safety issue with StringBuilder
is that method calls on a StringBuilder
do not synchronize.
Consider the implementation of the StringBuilder.append(char)
method:
public StringBuilder append(boolean b) {
super.append(b);
return this;
}
// from the superclass
public AbstractStringBuilder append(char c) {
int newCount = count + 1;
if (newCount > value.length)
expandCapacity(newCount);
value[count++] = c;
return this;
}
Now suppose that you have two thread that are sharing a StringBuilder
instance, and both attempt to append a character at the same time. Suppose that they both get to the value[count++] = c;
statement at the same time, and that count
is 1
. Each one will write its character into the buffer at value[1]
, and then update count
. Obviously only one character can be stored there ... so the other one will be lost. In addition, one of the increments to count
will probably be lost.
Worse than that, the value[count++] = c;
statement can fail even if the two threads don't arrive there at the same time. The reason is that the Java memory model says that unless there is proper synchronization (or more precisely, a happens before relationship), it is not guaranteed that the second thread will see the memory updates made by the first thread. What actually happens depends on whether and when the first thread's updates are written through to main memory.
Now lets look at StringBuffer.append(char)
:
public synchronized StringBuffer append(char c) {
super.append(c); // calls the "AbstractStringBuilder.append" method above.
return this;
}
Here we see that the append
method is synchronized
. This means two things:
Two threads cannot execute the superclass append
method on the same StringBuffer
object at the same time. Thus the first scenario cannot happen.
The synchronize
means that there is a happens before between successive calls to StringBuffer.append
made by different threads. That means that the later thread is guaranteed to see the updates made in the earlier one.
The String
case is different again. If we examine the code, we will see that there is no overt synchronization. But that's OK, because a String
object is effectively immutable; i.e are no methods in the String
API that will result in an externally observable change in the String
object's state. In addition:
The special behaviour of final
instance variables and constructors means that the all threads will see the correct initial state for any String
.
In the one place where the String
is mutable behind the scenes, the hashCode()
method will work correctly whether or not a thread sees the most recent changes to the hash
variable.
References: